wf-psf currently uses RectifiedAdam from TensorFlow Addons (TFA), which isn’t supported in the EDEN 3.1 (TensorFlow 2.11) environment.
We’ll remove the TFA dependency and make the optimizer configurable so users can still select it manually if desired.

Tasks

• Remove tensorflow-addons from required dependencies
• Add config-driven optimizer selection (Adam default, optional RectifiedAdam)
• Dynamically import TFA only if explicitly requested
• Replace TFA spline interpolation
• Update documentation and config templates
• Validate reproducibility on simulated datasets vs baseline

example config:

optimizer:
  name: adam            # or rectified_adam
  learning_rate: 0.001
  use_tensorflow_addons: false

Acceptance criteria

• Runs with TensorFlow 2.11 and no Addons installed
• Optional RectifiedAdam works when TFA is manually installed
• Training results consistent with baseline metrics